This paper discusses how criminal violence affects voting behavior and citizens’ demand for security policies on unequal and violent societies. I propose a theory considering both the micro-level dynamics behind preferences for security policies, and the supply of politicians framing the menu of security policies available to voters. I argue that, rather than priming valence consideration, security policies work as a wedge issue in which voters’ security preferences overlap with prior partisan identities and income status, as the salience of violence increases. Using the Brazilian case, one of the most violent countries in the world, I apply a combination of fine-grained observational data on crime and voting, computational text analysis on thousands of congressional speeches, and a novel factorial experiment to support my theory. Observational results show that crime shocks increase law-and-order candidates’ vote share, specially on more conservative municipalities. Within each city, the greater electoral support comes, particularly, from wealthier neighborhoods. Similar results are replicated using a factorial experiment on an online sample of Brazilian voters.